RFC 3986 percent-coding as its own extdeps model (re-derives #13390)

[gunbai-bot]

Re-derives gunbc#13390 (history only; its head is not a base) on current main.

What changed

• extdeps.uri keeps what it always was: the scheme/Uri value. RFC 3986 section 2.1 percent-coding is now its own model, extdeps.uri.percent_encoding, which cites the RFC through an ExternalAuthority anchor (DESIGN 3: cite the real upstream, model what it specifies). It owns the encode and decode folds, UriValidatedScalar (sole_constructor) and the typed refusals.
• The model uses the one UTF-8 encoder and decoder (std.bytes utf8_encode_bytes, std.encoding utf8_decode_octets) instead of the local fold extdeps.uri carried (single authority, DESIGN 2/3). The old comment that std.encoding sits above extdeps.uri is no longer true of the split.
• Importers repointed: github/app, namecheap/client, standards/rfc_8118, citation/pdf_safe_profile, principal_mention, roadmap_auth_routes, oracle_oci instance_ensure and tenancy_usage, the citation_cit1 and consumer witnesses, the guarantee-probe sole_constructor sources, the uri_percent encode/decode witnesses. Citations of extdeps.uri UriValidatedScalar now name the new module.
• std.os.invocation (root cause, section 6b): it used NonEmptyStr, List, CallingConvention with no imports. The larger closure the split pulls in (through std.encoding -> architecture_profile) made the census harness (type_ref_hit_ne_bind_measure, census-only) report a blocking UnresolvedType, which broke the sole_constructor probe claims. Added the three imports; no widening.
• Stage0 mirror extdeps_uri.rs regenerated with claim_executor --required-regen (only that surface drifted; planned=executed=adjudicated=169).

Receipts

• claim_batch on a seed built from this head: uri_percent_encode 13/13, uri_percent_decode 8/8, citation_cit1 22/22, citation_cit1_consumer 17/17, guarantee_probe_corpus 49 pass; its 4 failures (bare_none_generic_parameter_field_is_outside_the_wall_boundary, two sole_constructor green controls, dark_suite_dispositions_cover_migrated_v1_probes) fail identically on a clean main checkout with the same binary (pre-existing, unrelated to this change).
• gunbc test //gunbc/instruments:self-host exit 0 and //gunbc/instruments:v2-native-cli exit 0 at head 24c6400. srv1 was not reachable by ssh from this container (no identity), so both ran in this container, whose cgroup memory.max is 24 GiB (no systemd-run available here).

Not carried over

The Unicode from_code_point and git-fixture companions (#13378, #13392) are out of scope, per the brief.

🤖 Generated with Claude Code

Handoff (successor can continue from origin alone)

State: green on its head; landing packet is owed once #13711 lands. Next: on the parent's go, merge main ONCE by merge commit (no rebase), regen stage0 mirrors to the fixed point (claim_executor --source-root dag --source-root src/v2 --required-regen, RC 0; take conflicted mirrors from main), run the drift gate (generated_artifact_gate main_wet; revert unrelated sudoers drift), rerun claims and both instruments on srv1 (systemd-run --user --scope -p MemoryMax=22G -p MemorySwapMax=0), wait for witnesses + fresh review, send packet. Traps: delete gitignored roster.dag files (recurring_failure_mode, rung_drop) before each run; StaleBinary means rebuild after the last edit; do not merge main unless GitHub says DIRTY. Follow-up: half 2 of the from_code_point row (git fixture over percent-coding) on branch royal-owl-365/from-code-point after this lands. Notes: docs/handoff/royal-owl-365-from-code-point.md on that branch.

[gunbai-bot]

Review 78528: fixed in 25633e3. Reserved-scalar encode now takes its octets from std.bytes utf8_encode_bytes (bytes_octets fold into %HH via uri_percent_encode_octet_step); the lead/continuation math and the two/three/four-octet helpers are gone, so UTF-8 has one authority on both halves. Claims: encode 13/13, decode 8/8, citation_cit1 all PASS; self-host rc=0, v2-native-cli rc=0 on that head; required-regen drifts only in the pre-existing ietf mirror.
Review 78532: the stage0 extdeps_uri.rs is shrunk (-1163 lines), not deleted; the regen candidate matches it, so crate::extdeps_uri users resolve. No change. — sent from royal-owl-365

[gunbai-bot]

Superseded by #13730 (same tree plus main 1ed90d9, reopened from an id-free branch so the lane stays reachable after merge). Closed unmerged.